激光与光电子学进展, 2019, 56 (8): 081501, 网络出版: 2019-07-26
基于深度卷积神经网络的道路场景深度估计 下载: 1675次
Road Scene Depth Estimation Based on Deep Convolutional Neural Networks
机器视觉 深度卷积神经网路 深度估计 单目图像 深度学习 machine vision deep convolutional neural network depth estimation monocular image deep learning
摘要
提出了一种基于深度卷积神经网络的单目视觉深度估计方法,该方法采用端到端学习框架来构建模型。采用残差网络(ResNet)作为神经网络模型框架的编码部分来提取深度信息特征。采用密集连接卷积网络(DenseNet)对编码后的信息进行译码。通过Skip-Connections实现编码和解码的信息流的集成,避免了层间信息传输的丢失。实验结果表明,与其他单目视觉深度估计方法相比,使用深度卷积神经网络可以更有效准确地估计视觉深度。
Abstract
A monocular visual depth estimation method is proposed based on deep convolutional neural networks, in which an end-to-end learning framework is used to construct a model. A residual network (ResNet) is used as the coding part of the neural network model framework to extract the depth information features. The encoded information is decoded by a densely concatenated convolution network (DenseNet). The integration of the encoded and decoded information streams is realized by Skip-Connections, which avoids the loss of inter-layer information under transmission. The experimental results show that the depth convolution neural network can be used to estimate visual depth more effectively and accurately than other monocular visual depth estimation methods.
袁建中, 周武杰, 潘婷, 顾鹏笠. 基于深度卷积神经网络的道路场景深度估计[J]. 激光与光电子学进展, 2019, 56(8): 081501. Jianzhong Yuan, Wujie Zhou, Ting Pan, Pengli Gu. Road Scene Depth Estimation Based on Deep Convolutional Neural Networks[J]. Laser & Optoelectronics Progress, 2019, 56(8): 081501.